US 7106347 B1
In a passive pixel data handling system, pixel data may be transferred to a transfer function, at a given address range. The transfer function may perform a transformation and readdress the pixel data. For example, the data may be received through a media port target which transfers the pixel data to a transfer function located at an address range in virtual memory. Each transfer function may readdress the pixel data and forward it to a media port write back engine or to the memory address range of another transfer function.
1. A method comprising: transferring pixel data to a transformation engine at a given memory address range; performing a transformation on the pixel data; and readdressing the transformed pixel data to another memory address range of a second transformation engine using the transformation engine and without using a fetch engine.
2. The method of
manipulating the transformed pixel data without going between a memory location and another transformation engine.
3. The method of
writing pixel data to a first virtual memory location; and
performing a first pixel transformation at said first virtual memory location in a virtual memory space.
4. The method of
generating a virtual memory address for a second memory location.
5. The method of
re-mapping a virtual memory address of said first virtual memory location to write said transformed pixel data from said first virtual memory location to said virtual memory address of said second memory location; and
transferring the pixel data to a memory controller using a memory controller client in a forward, write-through direction.
6. The method of
7. The method of
8. The method of
9. The method of
10. The method of
11. The method of
12. The method of
13. An article comprising a medium storing instructions that enable a processor-based system to: transfer pixel data to a transformation engine at a given memory address range; perform a transformation on the pixel data; and readdress the transformed pixel data to another memory address range of a second transformation engine using the transformation engine and without using a fetch engine.
14. The article of
manipulate the transformed pixel data without going between a memory location and another transformation engine.
15. The article of
write pixel data to a first virtual memory location; and
perform a first pixel transformation at said first virtual memory location in a virtual memory space.
16. The article of
generate a virtual memory address for a second memory location.
17. The article of
re-map a virtual memory address of said first virtual memory location to write said transformed pixel data from said first virtual memory location to said virtual memory address of said second memory location; and
transfer the pixel data to a memory controller using a memory controller client in a forward write-through direction.
18. The article of
19. The article of
20. The article of
21. The article of
22. The article of
23. The article of
24. The article of
25. A system comprising:
a memory controller to receive pixel data and virtual memory addresses for a transformation of the pixel data in a virtual memory space;
a first memory controller client to forward the pixel data and virtual memory addresses to a first transfer function to perform the transformation of the pixel data; and
a second memory controller client to receive data from said first transfer function together with new virtual memory addresses for transfer in a forward, write-through direction without using a fetch engine.
26. The system of
27. The system of
28. The system of
29. The system of
This invention relates generally to graphics or video engines commonly used in computer systems.
A graphics or video engine is responsible for processing or manipulating data received from a pixel source such as a processor, a graphics controller or other devices. The graphics/video engine takes the pixel data and operates on that pixel data in a variety of well known ways. For example, the color space of the pixel data may be converted using routine algorithms. Alternatively, the pixel data may be scaled to change the size of the displayed information, again using well known algorithms. Similarly, the pixel data may be subjected to composition effects such as blurring, contrasting or other distortions.
In the conventional process, pixels are generated and deposited into memory as a result of a drawing operation such as three dimensional or video based rendering. To impose additional transformations like color space conversion or scaling, these pixels are then typically fetched from memory by a “fetch” engine. The operation is imposed and the pixels are then written back to the same memory location. The imposition of the transformation is termed “active” because it requires an explicit fetch engine to be set up with the parameters of the operation.
Thus, when a number of transformations are involved with given pixel data, many fetch engines may be needed that have the redundant functionality of generating pixel addresses. The use of these fetch engines complicates the memory controller that must operate with all of the fetch engines contending for memory bandwidth.
The need to use the fetch engines may also reduce the available memory bandwidth because of the need for read-modify-write cycles when imposing filtering operations after rendering. Moreover, when multiple transformations are needed, the programmer must awkwardly impose the transformations by causing the fetch engine to manipulate the data between a memory location and the various transformation engines. In addition, there is no easy way to cause multiple transformations to occur in a serial fashion.
Thus, there is a need for better ways to impose transformations on pixel data in graphic/video engines.
In accordance with one embodiment of the present invention, shown in
The pixel data operations in accordance with one embodiment of the present invention may be termed passive as compared to the active operations utilized in the prior art. An application may write pixels to a range of virtual memory addresses and the passive engine may impose the chosen operation. A new “re-mapped” memory address is generated and the pixel data is written to the new memory location.
Any conventional transformation or operation performed on pixel data including but not limited to, scaling, color conversion and composition can be implemented using a passive engine. Moreover, as indicated in
As shown in
The media port target 16 may provide the data to a variety of transfer functions 18 whose purpose is to receive pixel data and addresses from the media port target 16, perform a pixel and address transformation, and forward output pixel data and addresses to a media port write back engine 20. The media port write back engine 20 is a write-only memory controller client that receives pixel and address information from the individual transfer functions 18 and forwards the pixel data and addresses to the memory controller 14.
The operating system cooperates with the media port target 16 to set up a range of virtual memory addresses for use by the media port target 16. The operating system may set aside a specific memory range as virtual addresses that map to the media port target 16. Pixel data written to this memory range is not forwarded to physical memory. Instead, that pixel data is forwarded to the media port target 16. Each of the transfer functions 18 is also assigned a virtual memory range within the overall virtual memory range set aside for the media port target 16.
Thus, referring to
As a specific hypothetical example, the operating system may set aside the memory address range from 0x000000 to 0x800000 (8 megabytes) as virtual addresses that map to the media port target 16. The range of addresses from 0x000000 to 0x200000 may be set aside for a first transfer function 18 such as color conversion. The range of addresses from 0x200000 to 0x800000 may be set aside for a second transfer function 18. The color conversion function's output memory address range may be defined to be from 0x800000 through 0xA00000.
Each transfer function 18 performs a specific pixel transformation operation and address translation. The pixel data and addresses generated by the various transfer functions 18 may be written back to the memory controller 14. While a single memory port write back engine 20 is illustrated in
As another specific hypothetical example, the transfer function 18 a may be the color space transformation from the RGB color space to YUV color space. For example, the base address “baseinput” of the input pixel data may be 0x000000. The base address “baseoutput” of the transfer function 18 output pixel data may be 0x800000. Applying the transfer function 18, the RGB to YUV conversion operation may be as follows:
The addresses of the pixel data after color space conversion are transformed from their original pixel addresses. First, the offset to each pixel is determined by subtracting the baseinput from each pixel's original address. Then, the offset is added to the baseoutput to get the new address.
In this way, each RGB pixel data may be converted through the transformation equation indicated above to output YUV pixel data. The output YUV pixel data is deposited at memory location 0x800000 to 0XA00000 in an example in which the baseoutput is 0x800000 and the available memory range stops at 0XA00000. Similarly, the new addresses of the color space transformed pixel data is the offset added to the baseoutput to position the pixel data in easily determined locations, for the next transfer function or for return to memory.
As another hypothetical example, a first transfer function, such as a color space conversion, may be implemented in the virtual memory address range from 0x000000 to 0x200000 as indicated in
The first through third transfer functions may be performed serially by setting the output memory address range of the first transfer function to start at 0x200000, and setting the output memory address range of the second transfer function to start at 0x400000. Pixel data written to the range from 0x000000 to 0x200000 is color converted and forwarded to the virtual memory address range from 0x200000 through 0x400000. This pixel data is then scaled by horizontal averaging and written to the virtual memory address range from 0x400000 through 0x500000. The final output data of this sequence may written to the output memory address range designated by the third transfer function.
Transfer function software 32, shown in
An example of serialized transformations is shown in
In some embodiments of the present invention, there may be no need to use a plurality of fetch engines having redundant pixel address generating functionality. These fetch engines may complicate the memory controller that arbitrates among all of the engines contending for memory bandwidth. Transfer functions may be implemented, in accordance with embodiments of the present invention, under the same master memory controller although other implementations may also be possible. In addition, memory bandwidth utilization may be reduced by the elimination of the read-modify-write cycle typically used in active architectures when imposing filtering operations after rendering. In accordance with some embodiments of the present invention, pixel data being generated may be written directly to one or more successive memory-mapped media ports that perform the desired filter operations before returning the pixel data to memory.
In some embodiments of the present invention, multiple filter operations may be flexibly allocated and serially applied through intelligent memory aliasing. If a programmer wants to impose color space transformations, followed by scaling, followed by convolution operations, the programmer may do so by simply mapping one operation's output into the input of the next operation. With conventional active systems, filter operations are typically implemented in the middle of a fetch and write back circuit. If several operations are implemented and only one is used, the other operations are effectively unusable for the duration of the operation because of the use of a single, standalone execution pipe.
While the present invention has been described with respect to a limited number of embodiments, those skilled in the art will appreciate numerous modifications and variations therefrom. It is intended that the appended claims cover all such modifications and variations as fall within the true spirit and scope of this present invention.